Color Space Conversion

Every color in a digital image is stored as a combination of numbers. A screen mixes colors from red, green, and blue light, so it needs three numbers for these three components. A printing press, by contrast, works with ink on paper and mixes cyan, magenta, yellow, and black. The same red therefore has completely different numbers in the screen system than in the print system. Color space conversion is the calculation that translates these numbers into one another without the color shifting for the eye. Such a numbering system for colors is called a color space.

Why a photo looks different on every device

No device can display all the colors the human eye can see. Every device only covers a certain range, and these ranges differ greatly. A modern phone screen, for example, shows noticeably more saturated greens than a printing press. Without conversion, an image would appear vivid on one device and dull or distorted on another.

For companies, this is a real financial problem. If a furniture retailer shows a sofa in an online shop in a blue that the catalog print can’t match, returns come back. Brand colors are affected too: companies precisely specify which red or blue their logo must have. This color must be recognizably the same on posters, websites, and packaging.

A common misconception is that conversion is always lossless. That’s not true. Colors that don’t exist at all in the target system must be replaced by the nearest displayable color. This causes information to be lost, and converting back and forth repeatedly degrades an image step by step.

The detour via a neutral reference system

Calculating directly from one device to another would be impractical. With ten devices, you would need conversion rules for every possible pairing. Instead, a neutral intermediate system is used that isn’t tied to any particular device but describes how a human perceives color. The best-known one is called CIELAB. Each device then only needs to know the way into this intermediate system and back.

For this to work, every device needs a kind of profile. This file is called a color profile and records which real color corresponds to which combination of numbers. Well-known profiles for screens are sRGB and the larger Adobe RGB. A photo without an accompanying profile is like a measurement without a unit: the number alone doesn’t say enough.

For colors outside the target range, there are various strategies. One can compress all colors evenly so that the distances between them remain correct. Or one can leave the displayable colors unchanged and only push the extreme ones to the edge. Photos usually benefit from compression, because gradients stay smooth. For logos, the second approach is better, because the main color is hit exactly.

From phone cameras to AI image models

In everyday life, the conversion happens invisibly. When opening a photo, the operating system converts it to the connected monitor. When printing, the printer driver handles the conversion. Anyone who has ever received a printout that looked more matte than the preview has held the result of a color space conversion in their hands.

The process is also central in video technology. Streaming services often deliver movies in a format with especially bright and saturated colors, known as HDR for short. An older TV can’t display this and converts it down to its smaller color system. If this is done poorly, faces look pale or night scenes look muddy.

In AI development, conversion is a fixed step in data preparation. Images from the internet arrive in different color systems and are standardized before training, usually to sRGB. Otherwise, a model would learn random color shifts as a supposed pattern. Image generators output their results in a single, clearly defined color space for the same reason.